In the Northeast US, saturation excess is the most dominant runoff process and locations of runoff source areas, typically called variable source areas (VSAs), are determined by the available soil water storage and the landscape topographic position. To predict runoff generated from VSAs some water quality models use the Soil Conservation Service Curve Number equation (SCS-CN), which assumes a constant initial abstraction of rainfall is retained by the watershed prior to the beginning of runoff. We apply a VSA interpretation of the SCS-CN runoff equation that allows the initial abstraction to vary with antecedent moisture conditions. We couple this modified SCS-CN approach with a semi-distributed water balance model to predict runoff, and distribute predictions using a soil topographic index for the Town Brook watershed in the Catskill Mountains of New York State. The accuracy of predicted VSA extents using both the original and the modified SCS-CN equation were evaluated for 14 rainfall-runoff events through a comparison with average water table depths measured at 33 locations in Town Brook from March–September 2004. The modified SCS-CN equation captured VSA dynamics more accurately than the original equation. However, during events with high antecedent rainfall VSA dynamics were still under-predicted suggesting that VSA runoff is not captured solely by knowledge of the soil water deficit. Considering the importance of correctly predicting runoff generation and pollutant source areas in the landscape, the results of this study demonstrate the feasibility of integrating VSA hydrology into water quality models to reduce non-point source pollution. Copyright © 2009 John Wiley & Sons, Ltd.